Coconuts and coconut products are an important commodity in the Tongan economy. Plantations, such as the one in the town of Kolovai, have thousands of trees. Inventorying each of these trees by hand would require lots of time and manpower. Alternatively, tree health and location can be surveyed using remote sensing and deep learning. In this lesson, you'll use the Deep Learning tools in ArcGIS Pro to create training samples and run a deep learning model to identify the trees on the plantation. Then, you'll estimate tree health using a Visible Atmospherically Resistant Index (VARI) calculation to determine which trees may need inspection or maintenance.

To detect palm trees and calculate vegetation health, you only need ArcGIS Pro with the Image Analyst extension. To publish the palm tree health data as a feature service, you need ArcGIS Online and the Spatial Analyst extension.

In this lesson you will build skills in these areas:

• Creating training schema
• Digitizing training samples
• Using deep learning tools in ArcGIS Pro
• Calculating VARI
• Extracting data to points

Learn ArcGIS is a hands-on, problem-based learning website using real-world scenarios. Our mission is to encourage critical thinking, and to develop resources that support STEM education.

This resource contains flood extent maps of the Huallaga River near Chazuta, Peru derived from Height Above Nearest Drainage (HAND). MERIT Hydro global hydrography datasets (Hydrologically Adjusted Elevations, Flow Direction, and Upstream Drainage Pixel) were used as the input terrain data. The HAND raster and corresponding rating curve were creating using ArcHydro Pro tools in ArcGIS Pro. Specifically, this resource has a shapefile of flood extents for HAND values between 0-15, catchment and drainage line shapefiles, the HAND raster, and the rating curve as a csv. Using a flowrate, the HAND value can be determined from the rating curve. Then, the flood extent is the feature in the Chazuta_FloodExtent_HAND shapefile with the FloodValue attribute matching the Height value from the rating curve. As an example, this resource also includes a shapefile of the 5-m (Height value) flood extent along with the corresponding flood depth raster.

A 2D Hydraulic model (HEC-RAS) for below Tuttle Creek Reservoir at the confluence of the Kansas River and the Big Blue River near Manhattan, KS is presented. Model geometry is based on United States Geological Survey (USGS) 3DEP data (2015), with underwater bathymetry “burned” in using cross-sections sampled in the field in April of 2023. The model was calibrated based on water surface measured during data collection. The hydraulic simulations correspond to streamflows during which fish monitoring data were collected by researchers at Kansas State University (L. Rowley and K. Gido, to be published). Results from the hydraulic model, coupled with a sediment transport model, will be used to study fish and macroinvertabrate ecological response to streamflow. Methods The following is a summary of data utilized for developing a bathymetric terrain for 2D hydraulic modeling using HEC-RAS. Data used for model calibration and validation is also discussed.

Available Data Cross-section elevation data were collected by the United States Army Corps of Engineers (USACE) Kansas City District at approximately 200-foot to 1000-foot increments at the confluence of the Big Blue River and the Kansas River near Manhattan, Kansas. The following equipment was used by two complete surveying teams: • Ohmex SonarMite single beam echo sounder SFX @ 200khz, • Ohmex SonarMite single beam echo sounder DFX @ 28kHz & 200kHZ, • Trimble R12i 0096 & 0098, • Trimble R8 1984 & 6282

The cross-section elevation data were collected by boat and supplemented by hand-carried, pole-mounted Trimbles on April 10 to 14, 2023. The USGS gage on the Big Blue River near Manhattan, KS (06887000) had an average discharge of 425 cfs during the field collection time period (Figure 1). A USGS gage downstream of the confluence, Kansas River at Wamego, KS (06887500) shows an average discharge of 780 cfs at the same time period (Figure 2).

Figure 1 (Refer to supplemental information file). USGS gage Big Blue R NR Manhattan, KS – 06887000 discharge data for the week of April 11, 2023 – April 15, 2023. The average flow was taken as 425 cfs.

Figure 2 (Refer to supplemental information file). USGS gage Kansas River at Wamego, KS (06887500) discharge data for the week of April 11, 2023 – April 15, 2023. The average flow was taken as 780 cfs. Wamego, KS is downstream of the Big Blue River and Kansas River confluence and represents combined flow for both tributaries.

Figure 3 (Refer to supplemental information file). Map of bathymetric cross-sections collected in April 2023 near Manhattan, KS. Arrows show flow direction. Inset is the data collection location relative to the state of Kansas.

Terrain The field data collection featured 56 cross-sections. HEC-RAS 6.3.1 was utilized to create a bathymetric surface by interpolating 1-D cross-sections, while a 1-m resolution USGS 3DEP terrain (2015) was used for the floodplain and surrounding areas. A more recent USGS 3DEP (2018) data was available but featured higher stream flow than the 2015 data collection and therefore, more of the channel was submerged. Overall, the difference between 2015 and 2018 had a mean deviation of ~0.04 feet, with a majority of the differences in the channel ranging between +/-0.5 feet. Islands in this reach are unvegetated and prone to movement, and therefore the exact channel form is uncertain. However, it is assumed that relative island areas are consistent throughout the reach, and 2015 LiDAR was used to delineate the most island area as possible.

To build the bathymetric terrain, a similar process as what was discussed in Harris et al. (2023), field collected data were imported into ArcGIS Pro 3.0 as a point shapefile. To preserve georeferencing, the point shapefile was segmented into groups of 3-4 cross-sections and these cross-sections were interpolated into mini-surfaces using the Inverse Distance Weighted (IDW) spatial analysis tool. These mini-surfaces were brought into HEC-RAS and cross-sections were drawn to intersect with these field surveyed locations. The 1-D cross-sections were then used to create a TIFF for the entire channel area. The 1D interpolation captures the channel centerline between measured cross-sections but meanders and channel widening may not be covered by the interpolated channel. The channel raster was broken into its component objects or “exploded”, in ArcGIS Pro using the Raster to Point tool. The points were then interpolated using the Inverse-Distance-Weighted interpolation tool (IDW). This creates a terrain that covers meanders and channel expansion while maintaining fidelity to the original channel raster.

Areas where the terrain was inundated at the time of LiDAR data collection are “flat” and referred to as a hydro-flattened surface. The Slope tool in ArcMap was used to delineate these hydro-flattened areas and a shapefile tracing unsubmerged islands was used. The IDW surface was clipped to the hydro-flattened extents and then mosaicked with the original 3DEP terrain to create a seamless bathymetric and topographic surface.

The field data collected in April 2023 (Figure 3) required supplemental information to cover a fish monitoring instance upstream of the bridge at Pillsbury Drive/177. In September 2021, the USACE Kansas City District collected sediment samples with XY-georeference and depth measurements. The LiDAR hydro-flattened surface was used to estimate the energy grade slope from the new cross-section to the recent field monitoring extents. The model scenario or “plan” on the April 2023 extents was run at a similar flow as was occurring in September 2021. The combination of water surface elevation at that flow (780 cfs), the energy grade slope in the 3DEP data and field measured depth in 2021 were used to estimate the elevation at the channel bed.

Land Cover Land cover was delineated using the Multi-Resolution Land Characteristic (MRLC) Consortium’s 2019 National Land Cover Data (NLCD) (MRLC 2016). Fifteen types of landcover were identified for this study area by the NLCD: Hay-Pasture, Shrub-Scrub, Developed Low Intensity, Developed Medium Intensity, Cultivated Crops, Deciduous Forest, Herbaceous, Develop Open Space, Developed High Intensity, Woody Wetlands, Emergent Herbaceous Wetland, Open Water, Mixed Forest, Barren Land, and Evergreen Forest. Manning’s n values were selected based on a range of n values along with a “Suggested Initial n” provided by Krest Engineers (2021) (Table 1). Table 1. A table representing a range of Manning’s n values, a suggested Manning’s n value, and percent imperviousness for each NLCD land cover type. (Krest Engineers, 2021)

Model Settings The 2D HEC-RAS mesh was set to 40-feet square, with breaklines to orient cell edges along areas of steep elevation change or to support model convergence. Boundary conditions were placed at three locations in the 2D flow area: the inflow of the Big Blue River (boundary condition type: flow hydrograph), the upstream end of the Kanas River (flow hydrograph), and the downstream end of the Kanas River (normal depth). An energy grade slope was given as 0.0005 ft/ft for the Big Blue River and 0.0003 ft/ft for the Kansas River. Advanced time step control adjustments were implemented using Courant’s Criterion, with a minimum Courant of 0.75 and a maximum of 3.

Calibration The suggested value from Krest Engineers (2021) was the initial Manning’s n used for each land cover type (Table 1). The hydraulic model was then run, and the Manning’s n was changed to better conform to water surface elevations observed during field data collection. Flows corresponding to the field collection dates were 415 cfs for the Big Blue River and 360 cfs for the Kansas River. These streamflows were determined by cross-referencing the field collection dates (April 10 to 14, 2023) to continuous monitoring data available from USGS at gages Big Blue R NR Manhattan, KS (06887000) and Kansas R at Fort Riley, KS (06879100). The 2D model simulation results were compared to the field-measured water surface elevations at each channel cross-section with the ArcGIS Zonal Statistics as Table tool. Model improvement was determined by calculating the Root Mean Square Error (RMSE) of the simulated water surface elevation to the field observed water surface elevation, and the Manning’s n values resulting in the lowest error were selected. Following calibration, the model has overall RMSE of 0.29 ft for depth. The final Manning’s n values used for all the following simulations are included in Table 2.

Land Cover

Mannings n

Open Water

0.025

Emergent Herbaceous Wetlands

0.05

Woody Wetlands

0.045

Herbaceous

0.025

Mixed Forest

0.08

Evergreen Forest

0.08

Deciduous Forest

0.1

Scrub-Shrub

0.07

Hay-Pasture

0.025

Cultivated Crops

0.02

Baren Land

0.023

Developed, Open Space

0.03

Developed, Low Intensity

0.06

Developed, Medium Intensity

0.08

Developed, High Intensity

0.12

Table 2. The selected Manning’s n per Landcover classification after calibration

Simulations Apart from the calibration simulations, further simulations were conducted to match additional fish data collection from July 17 – 21, 2023 and October 2- 6, 2023. USGS gages, Big Blue R NR Manhattan, KS (06887000) and Kansas R at Fort Riley, KS (06879100), were used to find the discharge rates (in cfs) during those fish sampling periods. While discharge was consistent throughout the weeks for some gages (Figures 4 and 7), others showed differences greater than 10% or 100 cfs (Figures 5 and 6). The gages that showed significant differences were divided into two sub-simulations for the lower and higher flows during that week.

USGS Streamflow Data for July 17 - 21, 2023

HEC RAS Scenario Description River Simulation Flow (cfs)

July_KS_LF July lower flow Big

This map contains district-based ACT canopy cover polygons at 1m resolution (vegetation cover above 3m) as at April/May 2020, detected with Light Detection and Ranging (LiDAR). The full ACT vector dataset has been split into districts and 1km tiles to allow for easier viewing.WARNING: Due to complexity and high resolution this dataset may draw slowly or appear to miss some trees. For best results, zoom in, view single districts separately, or download a local copy.LIDAR data was acquired in April/May 2020 for the ACT under contract by Aerometrex, at an average resolution of 12ppm. LiDAR is classified to Level 3 (for ground) and delivered as LAS v1.4 in in GDA2020 MGA zone 55. Visit https://www.planning.act.gov.au/professionals/survey-spatial/spatial-information/lidar-data for more info. Processing was completed on the LAS GDA2020 MGA Zone 55 LAS tile set using AHD vertical datum at 1m resolution. Dataset represents high vegetation above 3m (tree canopy or tree-approximate objects - see caveats).Methodology:High noise errors were reclassified.Digital Surface Model (DSM) and Digital Elevation Model (DEM) surfaces were created using ArcPro 2.8 LAS Dataset Geoprocessing Tools at 1 m resolution.Canopy Height Model CHM was determined = Digital Surface Model (DSM) - Digital Elevation Model (DEM).Pits (empty cells inside tree canopies) of 2 pixel (2x1m) were removed using Nibble.An ACT Urban building footprint layer generated from the 2020 LIDAR dataset by Aerometrex was used to remove spurious canopy portions from building roof areas in the urban area.An NDVI layer from pan-sharpened Pleiades multispectral satellite imagery, acquired in April 2019 (the Uriarra area) and November 2019 (rest of Urban Area) was used to delete erroneous non-vegetative surfaces in urban areas.Canopy holes <=1m2 were filled by Eliminate tool.GDA2020 MGA Zone 55.Tree canopy lower than 3m were removed. Converted to a binary raster, then converted to vector.Other available datasets: ACT 2020 Canopy Height Model (raster), ACT Trees (Individual Delineated Trees with Height) (vector), DEM, DSM, Contours, Building Footprints (© Australian Capital Territory & Aerometrex Limited), ACT Permeability, shrub cover - contact spatialdata@act.gov.au. Full LAS Tile sets can also be obtained in the following vertical datums: GDA2020 ellipsoidal, AHD, AVWS - see https://elevation.fsdf.org.au/. Other derivative products (including other canopy products) are available on request or through the ACT Geospatial Data Catalogue.Road, Block and Division Canopy 2020 Statistics also available here: https://actmapi-actgov.opendata.arcgis.com/maps/act-canopy-cover-2020-statistics/aboutCaveats:1. Please note that most (if not all) CHM-based tree delineation results should be thought of as "tree-approximate objects", and not actual trees. 2. Although every effort was made to remove any erroneous polygons (such as street lights, back yard fences and powerlines), there are likely to be some errors remaining.Creative Commons by Attribution (CCBY) 4.0 (Australian Capital Territory). Any sharing, adaption/transformation and value adding, including commercial use should be attributed to ACT Government (Australian Capital Territory). This dataset has been created from the original LiDAR capture and classification © Australian Capital Territory & Aerometrex Limited 2020.How to cite this data: ACT Government (2020) (Botha, H). ACT Canopy Cover 1m 2020. City and Environment Directorate (CED), ACT Government. Canberra, ACT. Accessed via ACT Geospatial Data Catalogue.